Snitches may not get stitches, but they do get squeezed by higher rates. The tensions are high between Amazon and Anthropic, and they don’t appear to be easing any time soon. According to a new report from The Information, a contract between the companies that was renegotiated earlier this year will require Amazon to pay Anthropic based on token usage, leading to much higher bills for Amazon, and the online retail giant is looking for alternatives to Claude.
The new token-based pricing structure that Amazon will reportedly face does not kick in until next year, but its mere presence on the horizon could threaten a wide range of Amazon’s internal tools, given how heavily the company has leaned on Anthropic’s models. According to The Information, Amazon’s coding agent Kiro and workplace assistant Quick both rely on Claude. So does its consumer-facing Alexa for Shopping tool.
There is an obvious irony in Amazon reportedly bristling at token-based pricing after previously posting an internal leaderboard that encouraged employees to burn through as many tokens as possible. But the dispute may also point to a deeper rift in Amazon’s relationship with Anthropic.
That’s because the search for cheaper AI offerings has sent Amazon poking around a potential arrangement with Anthropic’s biggest rival, OpenAI. A deal between Amazon and OpenAI would move them even closer, and they have been getting mighty cozy lately. Earlier this year, the company committed $50 billion to the AI lab, with Amazon lending the company access to its infrastructure and OpenAI providing access to its models in return.
The closeness between OpenAI and Amazon comes as the e-commerce juggernaut has been growing apart from Anthropic, its first love. At the start of the AI “revolution,” Amazon put $4 billion into Anthropic’s coffers as OpenAI was in bed with Microsoft. And while the two have renewed their relationship multiple times, things have not been great between the companies lately.
For one, Anthropic has gotten big enough to draw unwanted attention, and it has started doing business with Amazon rivals, including a reportedly massive deal with Google for cloud services. But things really seemed to boil over last month when Anthropic released its Fable 5 AI model, the supposedly safe version of its “too powerful to release” Mythos model, only to have it clipped by the federal government after a report suggested it could go rogue.
That report came from Amazon, and happened to come as the company is reportedly ramping up to release its own cybersecurity-focused AI agent designed to spot vulnerabilities—the exact thing that Fable 5 is supposedly so good at.
